Hybrid Techniques for Diagnosing Endoscopy Images for Early Detection of Gastrointestinal Disease Based on Fusion Features

Table 4

ANN results with mixed features of CNN and handcrafted features for diagnostic endoscopy images of the gastroenterology dataset.

System	Disease types	Accuracy (%)	Sensitivity (%)	Precision (%)	Specificity (%)	AUC (%)
VGG-16 and handcrafted feature	Dyed_lifted_polyps	98.00	98.22	99.00	99.50	99.20
	Dyed_resection_margins	99.50	99.35	99.00	99.65	99.60
	Esophagitis	98.50	98.10	97.50	99.85	99.75
	Normal_cecum	98.50	97.85	98.50	99.75	99.35
	Normal_pylorus	98.50	98.40	99.00	99.45	99.50
	Normal_z_line	99.50	98.92	98.00	99.90	99.80
	Polyps	98.50	97.65	99.00	99.50	99.77
	Ulcerative_colitis	97.50	97.12	98.50	99.60	99.64
	Average ratio	98.60	98.20	98.56	99.65	99.58
DenseNet-121 and handcrafted feature	Dyed_lifted_polyps	98	98.25	98.50	99.56	99.63
	Dyed_resection_margins	99.5	99.20	99.00	100	99.60
	Esophagitis	98	97.80	99.50	99.49	99.38
	Normal_cecum	100	100.00	99.00	99.78	99.65
	Normal_pylorus	99.5	99.30	99.00	99.85	99.32
	Normal_z_line	99.5	99.44	98.50	99.90	99.57
	Polyps	98.5	97.78	99.50	99.60	99.10
	Ulcerative_colitis	98.50	97.85	98.50	100	99.80
	Average ratio	98.90	98.70	98.94	99.69	99.51